Evaluation and comparison of various deep neural networks for monocular depth estimation

Abstract

In this final year project, several testing scenarios and related methodology have been designed to examine the performance of the cutting-edge neural networks for monocular depth estimation. Since neural networks for monocular depth estimation is a fast-developing and emerging research field in recent years, neural network design and techniques involved keep evolving. It is both reasonable and beneficial to perceive different novel network design and implement these networks personally. If all the parameters during testing meet the lowest expectations in relative real-life application scenarios, it can be expected that neural networks will replace the dedicated depth sensors and make a huge difference in high-tech fields like artificial intelligence and autonomous driving.